Upload chat version of Qilin-Med-VL

config.json

@@ -0,0 +1,38 @@
+{
+    "architectures": [
+      "LlavaLlamaForCausalLM"
+    ],
+    "bos_token_id": 1,
+    "eos_token_id": 2,
+    "freeze_mm_mlp_adapter": false,
+    "hidden_act": "silu",
+    "hidden_size": 5120,
+    "image_aspect_ratio": "pad",
+    "image_grid_pinpoints": null,
+    "initializer_range": 0.02,
+    "intermediate_size": 13824,
+    "max_position_embeddings": 4096,
+    "mm_hidden_size": 1024,
+    "mm_projector_type": "mlp2x_gelu",
+    "mm_use_im_patch_token": false,
+    "mm_use_im_start_end": false,
+    "mm_vision_select_feature": "patch",
+    "mm_vision_select_layer": -2,
+    "mm_vision_tower": "openai/clip-vit-large-patch14-336",
+    "model_type": "llava",
+    "num_attention_heads": 40,
+    "num_hidden_layers": 40,
+    "num_key_value_heads": 40,
+    "pad_token_id": 0,
+    "pretraining_tp": 1,
+    "rms_norm_eps": 1e-05,
+    "rope_scaling": null,
+    "tie_word_embeddings": false,
+    "torch_dtype": "bfloat16",
+    "transformers_version": "4.31.0",
+    "tune_mm_mlp_adapter": false,
+    "use_cache": false,
+    "use_mm_proj": true,
+    "vocab_size": 32000
+  }
+  

pytorch_model-00001-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76ab07d425cb0d30dbfc16158c4634dcbddf3a1a2ffc6f9f4d380604bbecaad8
+size 9948726510

pytorch_model-00002-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1aa95282c696b1ecf5c5967234673439e9a5e5dd59da5a87eb14c4308bb5680a
+size 9904162976

pytorch_model-00003-of-00003.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8216e73d270d8934da5caa5778045c6f6f85626adb790717713790d42867afd1
+size 6241918849

pytorch_model.bin.index.json

@@ -0,0 +1,414 @@
+{
+  "metadata": {
+    "total_size": 26701028352
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00003-of-00003.bin",
+    "model.embed_tokens.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.15.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.15.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.2.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.20.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.input_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.mlp.down_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.3.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.30.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.30.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.30.mlp.up_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.30.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00003.bin",
+    "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00002-of-00003.bin",
+    "model.layers.31.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.32.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.33.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.34.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.35.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.36.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.37.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.38.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.input_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.mlp.down_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.mlp.gate_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.mlp.up_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.post_attention_layernorm.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.self_attn.k_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.self_attn.o_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.self_attn.q_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00003.bin",
+    "model.layers.39.self_attn.v_proj.weight": "pytorch_model-00003-of-00003.bin",
+    "model.layers.4.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.input_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.mlp.down_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.mlp.up_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00003.bin",
+    "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00001-of-00003.bin",
+    "model.mm_projector.0.bias": "pytorch_model-00003-of-00003.bin",
+    "model.mm_projector.0.weight": "pytorch_model-00003-of-00003.bin",
+    "model.mm_projector.2.bias": "pytorch_model-00003-of-00003.bin",
+    "model.mm_projector.2.weight": "pytorch_model-00003-of-00003.bin",
+    "model.norm.weight": "pytorch_model-00003-of-00003.bin"
+  }
+}

special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,35 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "legacy": false,
+  "model_max_length": 2048,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

trainer_state.json

@@ -0,0 +1,3316 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.523809523809524,
+  "global_step": 550,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.02,
+      "learning_rate": 1e-05,
+      "loss": 6.5129,
+      "step": 1
+    },
+    {
+      "epoch": 0.03,
+      "learning_rate": 2e-05,
+      "loss": 128.5625,
+      "step": 2
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 1.9983691039261358e-05,
+      "loss": 18.876,
+      "step": 3
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 1.99348173534855e-05,
+      "loss": 59.1932,
+      "step": 4
+    },
+    {
+      "epoch": 0.09,
+      "learning_rate": 1.9853538358476933e-05,
+      "loss": 379.3555,
+      "step": 5
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 1.9740119169423337e-05,
+      "loss": 4.746,
+      "step": 6
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 1.9594929736144978e-05,
+      "loss": 18.0314,
+      "step": 7
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 1.941844363639525e-05,
+      "loss": 38.1211,
+      "step": 8
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 1.92112365311485e-05,
+      "loss": 1059.6865,
+      "step": 9
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 1.8973984286913584e-05,
+      "loss": 256.4645,
+      "step": 10
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 1.8707460771197773e-05,
+      "loss": 79.4426,
+      "step": 11
+    },
+    {
+      "epoch": 0.21,
+      "learning_rate": 1.8412535328311813e-05,
+      "loss": 451.3965,
+      "step": 12
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 1.8090169943749477e-05,
+      "loss": 223.16,
+      "step": 13
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 1.7741416106390828e-05,
+      "loss": 263.6328,
+      "step": 14
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 1.736741137876405e-05,
+      "loss": 27.3438,
+      "step": 15
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 1.696937568655294e-05,
+      "loss": 2.4131,
+      "step": 16
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 1.6548607339452853e-05,
+      "loss": 34.8906,
+      "step": 17
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 1.6106478796354382e-05,
+      "loss": 0.5303,
+      "step": 18
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 1.5644432188667695e-05,
+      "loss": 53.5654,
+      "step": 19
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 1.5163974616389621e-05,
+      "loss": 409.5288,
+      "step": 20
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 1.4666673232256738e-05,
+      "loss": 411.0,
+      "step": 21
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 1.4154150130018867e-05,
+      "loss": 1.3561,
+      "step": 22
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 1.362807705350641e-05,
+      "loss": 222.3225,
+      "step": 23
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 1.3090169943749475e-05,
+      "loss": 74.2911,
+      "step": 24
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 1.2542183341934873e-05,
+      "loss": 86.5093,
+      "step": 25
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 1.1985904666457455e-05,
+      "loss": 562.082,
+      "step": 26
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 1.1423148382732854e-05,
+      "loss": 176.8262,
+      "step": 27
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 1.08557500847884e-05,
+      "loss": 416.7896,
+      "step": 28
+    },
+    {
+      "epoch": 0.5,
+      "learning_rate": 1.0285560507936962e-05,
+      "loss": 510.0781,
+      "step": 29
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 9.71443949206304e-06,
+      "loss": 133.375,
+      "step": 30
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 9.144249915211605e-06,
+      "loss": 2.4087,
+      "step": 31
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 8.576851617267151e-06,
+      "loss": 55.4609,
+      "step": 32
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 8.014095333542548e-06,
+      "loss": 217.9531,
+      "step": 33
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 7.4578166580651335e-06,
+      "loss": 35.3409,
+      "step": 34
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 6.909830056250527e-06,
+      "loss": 69.0106,
+      "step": 35
+    },
+    {
+      "epoch": 0.62,
+      "learning_rate": 6.3719229464935915e-06,
+      "loss": 0.0,
+      "step": 36
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 5.845849869981137e-06,
+      "loss": 199.4429,
+      "step": 37
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 5.333326767743263e-06,
+      "loss": 15.3613,
+      "step": 38
+    },
+    {
+      "epoch": 0.68,
+      "learning_rate": 4.836025383610382e-06,
+      "loss": 88.6268,
+      "step": 39
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 4.355567811332311e-06,
+      "loss": 73.5571,
+      "step": 40
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 3.893521203645618e-06,
+      "loss": 72.3561,
+      "step": 41
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 3.4513926605471504e-06,
+      "loss": 8.4178,
+      "step": 42
+    },
+    {
+      "epoch": 0.74,
+      "learning_rate": 3.0306243134470668e-06,
+      "loss": 20.0303,
+      "step": 43
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.6325886212359496e-06,
+      "loss": 2.5128,
+      "step": 44
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 2.2585838936091753e-06,
+      "loss": 7.6924,
+      "step": 45
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 1.9098300562505266e-06,
+      "loss": 73.0826,
+      "step": 46
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 1.587464671688187e-06,
+      "loss": 185.2307,
+      "step": 47
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 1.2925392288022299e-06,
+      "loss": 120.2432,
+      "step": 48
+    },
+    {
+      "epoch": 0.85,
+      "learning_rate": 1.0260157130864178e-06,
+      "loss": 13.2797,
+      "step": 49
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 7.887634688515e-07,
+      "loss": 14.5082,
+      "step": 50
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 1.111111111111111e-06,
+      "loss": 0.8499,
+      "step": 51
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 2.222222222222222e-06,
+      "loss": 0.8667,
+      "step": 52
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 3.3333333333333333e-06,
+      "loss": 0.827,
+      "step": 53
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 4.444444444444444e-06,
+      "loss": 0.8205,
+      "step": 54
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 5.555555555555557e-06,
+      "loss": 0.7996,
+      "step": 55
+    },
+    {
+      "epoch": 0.97,
+      "learning_rate": 6.666666666666667e-06,
+      "loss": 0.8256,
+      "step": 56
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 7.77777777777778e-06,
+      "loss": 0.7938,
+      "step": 57
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 8.888888888888888e-06,
+      "loss": 0.7957,
+      "step": 58
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1e-05,
+      "loss": 0.7284,
+      "step": 59
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.1111111111111113e-05,
+      "loss": 0.7159,
+      "step": 60
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.2222222222222224e-05,
+      "loss": 0.7319,
+      "step": 61
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.3333333333333333e-05,
+      "loss": 0.718,
+      "step": 62
+    },
+    {
+      "epoch": 1.09,
+      "learning_rate": 1.4444444444444446e-05,
+      "loss": 0.7225,
+      "step": 63
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.555555555555556e-05,
+      "loss": 0.6975,
+      "step": 64
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.6666666666666667e-05,
+      "loss": 0.686,
+      "step": 65
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.7777777777777777e-05,
+      "loss": 0.6895,
+      "step": 66
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.888888888888889e-05,
+      "loss": 0.726,
+      "step": 67
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 2e-05,
+      "loss": 0.6859,
+      "step": 68
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.9999838046468693e-05,
+      "loss": 0.6762,
+      "step": 69
+    },
+    {
+      "epoch": 1.21,
+      "learning_rate": 1.9999352191120556e-05,
+      "loss": 0.6762,
+      "step": 70
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.9998542449692794e-05,
+      "loss": 0.6844,
+      "step": 71
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.9997408848413494e-05,
+      "loss": 0.6845,
+      "step": 72
+    },
+    {
+      "epoch": 1.26,
+      "learning_rate": 1.999595142400081e-05,
+      "loss": 0.6556,
+      "step": 73
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.999417022366174e-05,
+      "loss": 0.6594,
+      "step": 74
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.999206530509063e-05,
+      "loss": 0.643,
+      "step": 75
+    },
+    {
+      "epoch": 1.32,
+      "learning_rate": 1.9989636736467278e-05,
+      "loss": 0.6619,
+      "step": 76
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.998688459645473e-05,
+      "loss": 0.6278,
+      "step": 77
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.9983808974196752e-05,
+      "loss": 0.6406,
+      "step": 78
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.9980409969314917e-05,
+      "loss": 0.6401,
+      "step": 79
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.9976687691905394e-05,
+      "loss": 0.613,
+      "step": 80
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.997264226253538e-05,
+      "loss": 0.6566,
+      "step": 81
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.9968273812239185e-05,
+      "loss": 0.6214,
+      "step": 82
+    },
+    {
+      "epoch": 1.44,
+      "learning_rate": 1.9963582482514003e-05,
+      "loss": 0.6824,
+      "step": 83
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.9958568425315316e-05,
+      "loss": 0.6209,
+      "step": 84
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.9953231803051977e-05,
+      "loss": 0.6382,
+      "step": 85
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.994757278858095e-05,
+      "loss": 0.6292,
+      "step": 86
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.9941591565201712e-05,
+      "loss": 0.6537,
+      "step": 87
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.9935288326650314e-05,
+      "loss": 0.6451,
+      "step": 88
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.99286632770931e-05,
+      "loss": 0.6433,
+      "step": 89
+    },
+    {
+      "epoch": 1.56,
+      "learning_rate": 1.992171663112011e-05,
+      "loss": 0.6449,
+      "step": 90
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 1.9914448613738107e-05,
+      "loss": 0.6204,
+      "step": 91
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 1.9906859460363307e-05,
+      "loss": 0.6378,
+      "step": 92
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 1.9898949416813757e-05,
+      "loss": 0.616,
+      "step": 93
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 1.9890718739301346e-05,
+      "loss": 0.5997,
+      "step": 94
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.988216769442353e-05,
+      "loss": 0.6771,
+      "step": 95
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1.98732965591547e-05,
+      "loss": 0.6288,
+      "step": 96
+    },
+    {
+      "epoch": 1.68,
+      "learning_rate": 1.9864105620837182e-05,
+      "loss": 0.656,
+      "step": 97
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 1.9854595177171968e-05,
+      "loss": 0.6113,
+      "step": 98
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 1.9844765536209045e-05,
+      "loss": 0.6113,
+      "step": 99
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 1.9834617016337424e-05,
+      "loss": 0.6346,
+      "step": 100
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1.9824149946274827e-05,
+      "loss": 0.6125,
+      "step": 101
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 1.981336466505705e-05,
+      "loss": 0.617,
+      "step": 102
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1.980226152202697e-05,
+      "loss": 0.6278,
+      "step": 103
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 1.979084087682323e-05,
+      "loss": 0.619,
+      "step": 104
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 1.9779103099368596e-05,
+      "loss": 0.6252,
+      "step": 105
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 1.9767048569857963e-05,
+      "loss": 0.6185,
+      "step": 106
+    },
+    {
+      "epoch": 1.85,
+      "learning_rate": 1.9754677678746064e-05,
+      "loss": 0.6144,
+      "step": 107
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 1.9741990826734793e-05,
+      "loss": 0.6116,
+      "step": 108
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 1.972898842476025e-05,
+      "loss": 0.6229,
+      "step": 109
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 1.9715670893979416e-05,
+      "loss": 0.6029,
+      "step": 110
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 1.9702038665756522e-05,
+      "loss": 0.588,
+      "step": 111
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 1.9688092181649065e-05,
+      "loss": 0.6343,
+      "step": 112
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 1.967383189339352e-05,
+      "loss": 0.6044,
+      "step": 113
+    },
+    {
+      "epoch": 1.97,
+      "learning_rate": 1.9659258262890683e-05,
+      "loss": 0.5969,
+      "step": 114
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1.964437176219075e-05,
+      "loss": 0.6034,
+      "step": 115
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 1.9629172873477995e-05,
+      "loss": 0.5693,
+      "step": 116
+    },
+    {
+      "epoch": 2.03,
+      "learning_rate": 1.9613662089055148e-05,
+      "loss": 0.4744,
+      "step": 117
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 1.9597839911327475e-05,
+      "loss": 0.4942,
+      "step": 118
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 1.9581706852786492e-05,
+      "loss": 0.4684,
+      "step": 119
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 1.956526343599335e-05,
+      "loss": 0.4619,
+      "step": 120
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 1.9548510193561938e-05,
+      "loss": 0.4727,
+      "step": 121
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 1.953144766814161e-05,
+      "loss": 0.4564,
+      "step": 122
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 1.9514076412399615e-05,
+      "loss": 0.4622,
+      "step": 123
+    },
+    {
+      "epoch": 2.15,
+      "learning_rate": 1.9496396989003195e-05,
+      "loss": 0.4508,
+      "step": 124
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 1.947840997060136e-05,
+      "loss": 0.4685,
+      "step": 125
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 1.946011593980634e-05,
+      "loss": 0.439,
+      "step": 126
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 1.9441515489174708e-05,
+      "loss": 0.4212,
+      "step": 127
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 1.9422609221188208e-05,
+      "loss": 0.4541,
+      "step": 128
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 1.94033977482342e-05,
+      "loss": 0.4522,
+      "step": 129
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 1.938388169258587e-05,
+      "loss": 0.425,
+      "step": 130
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 1.9364061686382042e-05,
+      "loss": 0.4692,
+      "step": 131
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 1.9343938371606714e-05,
+      "loss": 0.458,
+      "step": 132
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 1.9323512400068262e-05,
+      "loss": 0.448,
+      "step": 133
+    },
+    {
+      "epoch": 2.32,
+      "learning_rate": 1.9302784433378333e-05,
+      "loss": 0.4646,
+      "step": 134
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 1.928175514293041e-05,
+      "loss": 0.4385,
+      "step": 135
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 1.9260425209878052e-05,
+      "loss": 0.457,
+      "step": 136
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 1.9238795325112867e-05,
+      "loss": 0.4599,
+      "step": 137
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 1.9216866189242095e-05,
+      "loss": 0.4387,
+      "step": 138
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1.9194638512565937e-05,
+      "loss": 0.4588,
+      "step": 139
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.917211301505453e-05,
+      "loss": 0.4611,
+      "step": 140
+    },
+    {
+      "epoch": 2.44,
+      "learning_rate": 1.9149290426324658e-05,
+      "loss": 0.4617,
+      "step": 141
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.912617148561608e-05,
+      "loss": 0.4652,
+      "step": 142
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.9102756941767625e-05,
+      "loss": 0.4507,
+      "step": 143
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.907904755319289e-05,
+      "loss": 0.4509,
+      "step": 144
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.9055044087855728e-05,
+      "loss": 0.4532,
+      "step": 145
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.903074732324533e-05,
+      "loss": 0.4551,
+      "step": 146
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.900615804635106e-05,
+      "loss": 0.4452,
+      "step": 147
+    },
+    {
+      "epoch": 2.56,
+      "learning_rate": 1.8981277053636963e-05,
+      "loss": 0.4509,
+      "step": 148
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.8956105151015966e-05,
+      "loss": 0.4544,
+      "step": 149
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 1.8930643153823777e-05,
+      "loss": 0.4436,
+      "step": 150
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 1.8904891886792465e-05,
+      "loss": 0.4397,
+      "step": 151
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 1.8878852184023754e-05,
+      "loss": 0.4456,
+      "step": 152
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 1.885252488896201e-05,
+      "loss": 0.4397,
+      "step": 153
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 1.8825910854366914e-05,
+      "loss": 0.4551,
+      "step": 154
+    },
+    {
+      "epoch": 2.68,
+      "learning_rate": 1.879901094228584e-05,
+      "loss": 0.4485,
+      "step": 155
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 1.8771826024025944e-05,
+      "loss": 0.451,
+      "step": 156
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 1.8744356980125922e-05,
+      "loss": 0.4481,
+      "step": 157
+    },
+    {
+      "epoch": 2.74,
+      "learning_rate": 1.8716604700327516e-05,
+      "loss": 0.4463,
+      "step": 158
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 1.8688570083546658e-05,
+      "loss": 0.4459,
+      "step": 159
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 1.866025403784439e-05,
+      "loss": 0.4586,
+      "step": 160
+    },
+    {
+      "epoch": 2.79,
+      "learning_rate": 1.863165748039743e-05,
+      "loss": 0.4512,
+      "step": 161
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 1.8602781337468472e-05,
+      "loss": 0.473,
+      "step": 162
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 1.857362654437618e-05,
+      "loss": 0.457,
+      "step": 163
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 1.8544194045464888e-05,
+      "loss": 0.4694,
+      "step": 164
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 1.8514484794074028e-05,
+      "loss": 0.4526,
+      "step": 165
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 1.8484499752507234e-05,
+      "loss": 0.4473,
+      "step": 166
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 1.845423989200118e-05,
+      "loss": 0.4438,
+      "step": 167
+    },
+    {
+      "epoch": 2.91,
+      "learning_rate": 1.8423706192694118e-05,
+      "loss": 0.4663,
+      "step": 168
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 1.8392899643594135e-05,
+      "loss": 0.4682,
+      "step": 169
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 1.836182124254711e-05,
+      "loss": 0.4545,
+      "step": 170
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 1.8330471996204408e-05,
+      "loss": 0.4507,
+      "step": 171
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 1.8298852919990254e-05,
+      "loss": 0.4735,
+      "step": 172
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 1.8266965038068856e-05,
+      "loss": 0.4431,
+      "step": 173
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 1.823480938331124e-05,
+      "loss": 0.3441,
+      "step": 174
+    },
+    {
+      "epoch": 3.03,
+      "learning_rate": 1.820238699726177e-05,
+      "loss": 0.3109,
+      "step": 175
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 1.816969893010442e-05,
+      "loss": 0.2899,
+      "step": 176
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 1.8136746240628783e-05,
+      "loss": 0.2861,
+      "step": 177
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 1.810352999619574e-05,
+      "loss": 0.2917,
+      "step": 178
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 1.8070051272702905e-05,
+      "loss": 0.2885,
+      "step": 179
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 1.8036311154549783e-05,
+      "loss": 0.2769,
+      "step": 180
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 1.8002310734602625e-05,
+      "loss": 0.2687,
+      "step": 181
+    },
+    {
+      "epoch": 3.15,
+      "learning_rate": 1.7968051114159046e-05,
+      "loss": 0.2703,
+      "step": 182
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 1.7933533402912354e-05,
+      "loss": 0.2642,
+      "step": 183
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 1.789875871891559e-05,
+      "loss": 0.2688,
+      "step": 184
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 1.7863728188545326e-05,
+      "loss": 0.2778,
+      "step": 185
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 1.7828442946465188e-05,
+      "loss": 0.2652,
+      "step": 186
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 1.7792904135589085e-05,
+      "loss": 0.2645,
+      "step": 187
+    },
+    {
+      "epoch": 3.26,
+      "learning_rate": 1.77571129070442e-05,
+      "loss": 0.2801,
+      "step": 188
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 1.7721070420133702e-05,
+      "loss": 0.2712,
+      "step": 189
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 1.7684777842299206e-05,
+      "loss": 0.2563,
+      "step": 190
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 1.7648236349082928e-05,
+      "loss": 0.2657,
+      "step": 191
+    },
+    {
+      "epoch": 3.32,
+      "learning_rate": 1.761144712408965e-05,
+      "loss": 0.2544,
+      "step": 192
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 1.7574411358948347e-05,
+      "loss": 0.2623,
+      "step": 193
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 1.7537130253273613e-05,
+      "loss": 0.2522,
+      "step": 194
+    },
+    {
+      "epoch": 3.38,
+      "learning_rate": 1.7499605014626786e-05,
+      "loss": 0.2576,
+      "step": 195
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 1.7461836858476858e-05,
+      "loss": 0.2722,
+      "step": 196
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 1.742382700816107e-05,
+      "loss": 0.2766,
+      "step": 197
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 1.7385576694845324e-05,
+      "loss": 0.2663,
+      "step": 198
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 1.7347087157484282e-05,
+      "loss": 0.2778,
+      "step": 199
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 1.730835964278124e-05,
+      "loss": 0.2678,
+      "step": 200
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 1.726939540514776e-05,
+      "loss": 0.2769,
+      "step": 201
+    },
+    {
+      "epoch": 3.5,
+      "learning_rate": 1.723019570666301e-05,
+      "loss": 0.2662,
+      "step": 202
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 1.719076181703291e-05,
+      "loss": 0.2684,
+      "step": 203
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 1.7151095013548996e-05,
+      "loss": 0.2758,
+      "step": 204
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 1.711119658104705e-05,
+      "loss": 0.2668,
+      "step": 205
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 1.7071067811865477e-05,
+      "loss": 0.2708,
+      "step": 206
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 1.7030710005803453e-05,
+      "loss": 0.2663,
+      "step": 207
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 1.699012447007882e-05,
+      "loss": 0.2744,
+      "step": 208
+    },
+    {
+      "epoch": 3.62,
+      "learning_rate": 1.694931251928575e-05,
+      "loss": 0.2821,
+      "step": 209
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 1.690827547535214e-05,
+      "loss": 0.2635,
+      "step": 210
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 1.6867014667496838e-05,
+      "loss": 0.2729,
+      "step": 211
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 1.6825531432186545e-05,
+      "loss": 0.2759,
+      "step": 212
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 1.6783827113092547e-05,
+      "loss": 0.2795,
+      "step": 213
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 1.6741903061047204e-05,
+      "loss": 0.2857,
+      "step": 214
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 1.6699760634000166e-05,
+      "loss": 0.2833,
+      "step": 215
+    },
+    {
+      "epoch": 3.74,
+      "learning_rate": 1.6657401196974405e-05,
+      "loss": 0.2844,
+      "step": 216
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 1.6614826122022015e-05,
+      "loss": 0.2868,
+      "step": 217
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 1.6572036788179728e-05,
+      "loss": 0.2819,
+      "step": 218
+    },
+    {
+      "epoch": 3.79,
+      "learning_rate": 1.6529034581424293e-05,
+      "loss": 0.2739,
+      "step": 219
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 1.648582089462756e-05,
+      "loss": 0.2735,
+      "step": 220
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 1.6442397127511366e-05,
+      "loss": 0.2718,
+      "step": 221
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 1.6398764686602188e-05,
+      "loss": 0.2751,
+      "step": 222
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 1.6354924985185614e-05,
+      "loss": 0.2729,
+      "step": 223
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 1.631087944326053e-05,
+      "loss": 0.2758,
+      "step": 224
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 1.6266629487493144e-05,
+      "loss": 0.27,
+      "step": 225
+    },
+    {
+      "epoch": 3.91,
+      "learning_rate": 1.622217655117078e-05,
+      "loss": 0.2755,
+      "step": 226
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 1.6177522074155436e-05,
+      "loss": 0.2919,
+      "step": 227
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 1.6132667502837164e-05,
+      "loss": 0.2715,
+      "step": 228
+    },
+    {
+      "epoch": 3.97,
+      "learning_rate": 1.608761429008721e-05,
+      "loss": 0.2761,
+      "step": 229
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 1.6042363895210948e-05,
+      "loss": 0.2746,
+      "step": 230
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 1.5996917783900633e-05,
+      "loss": 0.2849,
+      "step": 231
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 1.59512774281879e-05,
+      "loss": 0.1437,
+      "step": 232
+    },
+    {
+      "epoch": 4.03,
+      "learning_rate": 1.590544430639611e-05,
+      "loss": 0.1407,
+      "step": 233
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 1.585941990309245e-05,
+      "loss": 0.138,
+      "step": 234
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 1.5813205709039842e-05,
+      "loss": 0.1311,
+      "step": 235
+    },
+    {
+      "epoch": 4.09,
+      "learning_rate": 1.5766803221148676e-05,
+      "loss": 0.1281,
+      "step": 236
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 1.57202139424283e-05,
+      "loss": 0.1323,
+      "step": 237
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 1.5673439381938365e-05,
+      "loss": 0.129,
+      "step": 238
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 1.5626481054739916e-05,
+      "loss": 0.1325,
+      "step": 239
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 1.5579340481846338e-05,
+      "loss": 0.1222,
+      "step": 240
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 1.5532019190174074e-05,
+      "loss": 0.1305,
+      "step": 241
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 1.5484518712493188e-05,
+      "loss": 0.1165,
+      "step": 242
+    },
+    {
+      "epoch": 4.21,
+      "learning_rate": 1.54368405873777e-05,
+      "loss": 0.1212,
+      "step": 243
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 1.538898635915576e-05,
+      "loss": 0.1282,
+      "step": 244
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 1.5340957577859605e-05,
+      "loss": 0.1267,
+      "step": 245
+    },
+    {
+      "epoch": 4.26,
+      "learning_rate": 1.529275579917539e-05,
+      "loss": 0.1216,
+      "step": 246
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 1.5244382584392772e-05,
+      "loss": 0.1196,
+      "step": 247
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 1.5195839500354337e-05,
+      "loss": 0.1202,
+      "step": 248
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 1.5147128119404863e-05,
+      "loss": 0.1272,
+      "step": 249
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 1.5098250019340385e-05,
+      "loss": 0.1207,
+      "step": 250
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 1.5049206783357082e-05,
+      "loss": 0.1233,
+      "step": 251
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 1.5000000000000002e-05,
+      "loss": 0.1124,
+      "step": 252
+    },
+    {
+      "epoch": 4.38,
+      "learning_rate": 1.4950631263111615e-05,
+      "loss": 0.1131,
+      "step": 253
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 1.4901102171780175e-05,
+      "loss": 0.1165,
+      "step": 254
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 1.485141433028793e-05,
+      "loss": 0.1198,
+      "step": 255
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 1.4801569348059158e-05,
+      "loss": 0.1199,
+      "step": 256
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 1.4751568839608036e-05,
+      "loss": 0.1171,
+      "step": 257
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 1.4701414424486353e-05,
+      "loss": 0.12,
+      "step": 258
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 1.4651107727231032e-05,
+      "loss": 0.1239,
+      "step": 259
+    },
+    {
+      "epoch": 4.5,
+      "learning_rate": 1.4600650377311523e-05,
+      "loss": 0.1157,
+      "step": 260
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 1.455004400907703e-05,
+      "loss": 0.1188,
+      "step": 261
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 1.4499290261703565e-05,
+      "loss": 0.1143,
+      "step": 262
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 1.4448390779140844e-05,
+      "loss": 0.1203,
+      "step": 263
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 1.4397347210059059e-05,
+      "loss": 0.1164,
+      "step": 264
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 1.4346161207795464e-05,
+      "loss": 0.1323,
+      "step": 265
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 1.4294834430300822e-05,
+      "loss": 0.1238,
+      "step": 266
+    },
+    {
+      "epoch": 4.62,
+      "learning_rate": 1.4243368540085702e-05,
+      "loss": 0.1254,
+      "step": 267
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 1.4191765204166643e-05,
+      "loss": 0.1245,
+      "step": 268
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 1.4140026094012136e-05,
+      "loss": 0.1255,
+      "step": 269
+    },
+    {
+      "epoch": 4.68,
+      "learning_rate": 1.4088152885488504e-05,
+      "loss": 0.1165,
+      "step": 270
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 1.4036147258805604e-05,
+      "loss": 0.125,
+      "step": 271
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 1.3984010898462417e-05,
+      "loss": 0.1209,
+      "step": 272
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 1.3931745493192473e-05,
+      "loss": 0.1205,
+      "step": 273
+    },
+    {
+      "epoch": 4.74,
+      "learning_rate": 1.3879352735909163e-05,
+      "loss": 0.1231,
+      "step": 274
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 1.3826834323650899e-05,
+      "loss": 0.1197,
+      "step": 275
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 1.3774191957526144e-05,
+      "loss": 0.1191,
+      "step": 276
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 1.3721427342658322e-05,
+      "loss": 0.1205,
+      "step": 277
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 1.3668542188130567e-05,
+      "loss": 0.1185,
+      "step": 278
+    },
+    {
+      "epoch": 4.83,
+      "learning_rate": 1.3615538206930387e-05,
+      "loss": 0.1214,
+      "step": 279
+    },
+    {
+      "epoch": 4.85,
+      "learning_rate": 1.356241711589417e-05,
+      "loss": 0.1187,
+      "step": 280
+    },
+    {
+      "epoch": 4.87,
+      "learning_rate": 1.350918063565157e-05,
+      "loss": 0.1358,
+      "step": 281
+    },
+    {
+      "epoch": 4.88,
+      "learning_rate": 1.3455830490569782e-05,
+      "loss": 0.1283,
+      "step": 282
+    },
+    {
+      "epoch": 4.9,
+      "learning_rate": 1.3402368408697681e-05,
+      "loss": 0.1251,
+      "step": 283
+    },
+    {
+      "epoch": 4.92,
+      "learning_rate": 1.3348796121709862e-05,
+      "loss": 0.1201,
+      "step": 284
+    },
+    {
+      "epoch": 4.94,
+      "learning_rate": 1.3295115364850535e-05,
+      "loss": 0.1292,
+      "step": 285
+    },
+    {
+      "epoch": 4.95,
+      "learning_rate": 1.3241327876877328e-05,
+      "loss": 0.1245,
+      "step": 286
+    },
+    {
+      "epoch": 4.97,
+      "learning_rate": 1.318743540000496e-05,
+      "loss": 0.127,
+      "step": 287
+    },
+    {
+      "epoch": 4.99,
+      "learning_rate": 1.3133439679848824e-05,
+      "loss": 0.1168,
+      "step": 288
+    },
+    {
+      "epoch": 5.0,
+      "learning_rate": 1.307934246536843e-05,
+      "loss": 0.115,
+      "step": 289
+    },
+    {
+      "epoch": 5.02,
+      "learning_rate": 1.302514550881076e-05,
+      "loss": 0.0562,
+      "step": 290
+    },
+    {
+      "epoch": 5.04,
+      "learning_rate": 1.2970850565653515e-05,
+      "loss": 0.0558,
+      "step": 291
+    },
+    {
+      "epoch": 5.06,
+      "learning_rate": 1.291645939454825e-05,
+      "loss": 0.0531,
+      "step": 292
+    },
+    {
+      "epoch": 5.07,
+      "learning_rate": 1.2861973757263416e-05,
+      "loss": 0.0496,
+      "step": 293
+    },
+    {
+      "epoch": 5.09,
+      "learning_rate": 1.2807395418627278e-05,
+      "loss": 0.0512,
+      "step": 294
+    },
+    {
+      "epoch": 5.11,
+      "learning_rate": 1.2752726146470775e-05,
+      "loss": 0.0507,
+      "step": 295
+    },
+    {
+      "epoch": 5.13,
+      "learning_rate": 1.2697967711570243e-05,
+      "loss": 0.0527,
+      "step": 296
+    },
+    {
+      "epoch": 5.14,
+      "learning_rate": 1.2643121887590064e-05,
+      "loss": 0.0497,
+      "step": 297
+    },
+    {
+      "epoch": 5.16,
+      "learning_rate": 1.2588190451025209e-05,
+      "loss": 0.0467,
+      "step": 298
+    },
+    {
+      "epoch": 5.18,
+      "learning_rate": 1.2533175181143704e-05,
+      "loss": 0.0483,
+      "step": 299
+    },
+    {
+      "epoch": 5.19,
+      "learning_rate": 1.2478077859929e-05,
+      "loss": 0.0485,
+      "step": 300
+    },
+    {
+      "epoch": 5.21,
+      "learning_rate": 1.2422900272022242e-05,
+      "loss": 0.0511,
+      "step": 301
+    },
+    {
+      "epoch": 5.23,
+      "learning_rate": 1.2367644204664468e-05,
+      "loss": 0.051,
+      "step": 302
+    },
+    {
+      "epoch": 5.25,
+      "learning_rate": 1.2312311447638731e-05,
+      "loss": 0.0477,
+      "step": 303
+    },
+    {
+      "epoch": 5.26,
+      "learning_rate": 1.2256903793212107e-05,
+      "loss": 0.0435,
+      "step": 304
+    },
+    {
+      "epoch": 5.28,
+      "learning_rate": 1.2201423036077657e-05,
+      "loss": 0.0508,
+      "step": 305
+    },
+    {
+      "epoch": 5.3,
+      "learning_rate": 1.2145870973296288e-05,
+      "loss": 0.0486,
+      "step": 306
+    },
+    {
+      "epoch": 5.32,
+      "learning_rate": 1.2090249404238548e-05,
+      "loss": 0.0475,
+      "step": 307
+    },
+    {
+      "epoch": 5.33,
+      "learning_rate": 1.2034560130526341e-05,
+      "loss": 0.0465,
+      "step": 308
+    },
+    {
+      "epoch": 5.35,
+      "learning_rate": 1.1978804955974573e-05,
+      "loss": 0.0464,
+      "step": 309
+    },
+    {
+      "epoch": 5.37,
+      "learning_rate": 1.1922985686532726e-05,
+      "loss": 0.0467,
+      "step": 310
+    },
+    {
+      "epoch": 5.39,
+      "learning_rate": 1.1867104130226363e-05,
+      "loss": 0.0462,
+      "step": 311
+    },
+    {
+      "epoch": 5.4,
+      "learning_rate": 1.1811162097098559e-05,
+      "loss": 0.0499,
+      "step": 312
+    },
+    {
+      "epoch": 5.42,
+      "learning_rate": 1.1755161399151277e-05,
+      "loss": 0.0515,
+      "step": 313
+    },
+    {
+      "epoch": 5.44,
+      "learning_rate": 1.1699103850286668e-05,
+      "loss": 0.0473,
+      "step": 314
+    },
+    {
+      "epoch": 5.45,
+      "learning_rate": 1.1642991266248338e-05,
+      "loss": 0.0467,
+      "step": 315
+    },
+    {
+      "epoch": 5.47,
+      "learning_rate": 1.1586825464562515e-05,
+      "loss": 0.0452,
+      "step": 316
+    },
+    {
+      "epoch": 5.49,
+      "learning_rate": 1.153060826447918e-05,
+      "loss": 0.0457,
+      "step": 317
+    },
+    {
+      "epoch": 5.51,
+      "learning_rate": 1.1474341486913146e-05,
+      "loss": 0.0459,
+      "step": 318
+    },
+    {
+      "epoch": 5.52,
+      "learning_rate": 1.1418026954385082e-05,
+      "loss": 0.0463,
+      "step": 319
+    },
+    {
+      "epoch": 5.54,
+      "learning_rate": 1.1361666490962468e-05,
+      "loss": 0.0453,
+      "step": 320
+    },
+    {
+      "epoch": 5.56,
+      "learning_rate": 1.130526192220052e-05,
+      "loss": 0.0497,
+      "step": 321
+    },
+    {
+      "epoch": 5.58,
+      "learning_rate": 1.1248815075083051e-05,
+      "loss": 0.0488,
+      "step": 322
+    },
+    {
+      "epoch": 5.59,
+      "learning_rate": 1.1192327777963313e-05,
+      "loss": 0.0444,
+      "step": 323
+    },
+    {
+      "epoch": 5.61,
+      "learning_rate": 1.113580186050475e-05,
+      "loss": 0.049,
+      "step": 324
+    },
+    {
+      "epoch": 5.63,
+      "learning_rate": 1.1079239153621753e-05,
+      "loss": 0.0412,
+      "step": 325
+    },
+    {
+      "epoch": 5.65,
+      "learning_rate": 1.1022641489420342e-05,
+      "loss": 0.0475,
+      "step": 326
+    },
+    {
+      "epoch": 5.66,
+      "learning_rate": 1.0966010701138841e-05,
+      "loss": 0.0462,
+      "step": 327
+    },
+    {
+      "epoch": 5.68,
+      "learning_rate": 1.0909348623088472e-05,
+      "loss": 0.0453,
+      "step": 328
+    },
+    {
+      "epoch": 5.7,
+      "learning_rate": 1.0852657090593961e-05,
+      "loss": 0.0469,
+      "step": 329
+    },
+    {
+      "epoch": 5.71,
+      "learning_rate": 1.0795937939934088e-05,
+      "loss": 0.0469,
+      "step": 330
+    },
+    {
+      "epoch": 5.73,
+      "learning_rate": 1.0739193008282203e-05,
+      "loss": 0.0446,
+      "step": 331
+    },
+    {
+      "epoch": 5.75,
+      "learning_rate": 1.0682424133646712e-05,
+      "loss": 0.0504,
+      "step": 332
+    },
+    {
+      "epoch": 5.77,
+      "learning_rate": 1.062563315481156e-05,
+      "loss": 0.0433,
+      "step": 333
+    },
+    {
+      "epoch": 5.78,
+      "learning_rate": 1.056882191127667e-05,
+      "loss": 0.0457,
+      "step": 334
+    },
+    {
+      "epoch": 5.8,
+      "learning_rate": 1.0511992243198335e-05,
+      "loss": 0.0474,
+      "step": 335
+    },
+    {
+      "epoch": 5.82,
+      "learning_rate": 1.0455145991329639e-05,
+      "loss": 0.0439,
+      "step": 336
+    },
+    {
+      "epoch": 5.84,
+      "learning_rate": 1.039828499696083e-05,
+      "loss": 0.0423,
+      "step": 337
+    },
+    {
+      "epoch": 5.85,
+      "learning_rate": 1.034141110185968e-05,
+      "loss": 0.0468,
+      "step": 338
+    },
+    {
+      "epoch": 5.87,
+      "learning_rate": 1.0284526148211815e-05,
+      "loss": 0.0448,
+      "step": 339
+    },
+    {
+      "epoch": 5.89,
+      "learning_rate": 1.0227631978561057e-05,
+      "loss": 0.047,
+      "step": 340
+    },
+    {
+      "epoch": 5.9,
+      "learning_rate": 1.017073043574975e-05,
+      "loss": 0.0458,
+      "step": 341
+    },
+    {
+      "epoch": 5.92,
+      "learning_rate": 1.0113823362859042e-05,
+      "loss": 0.0461,
+      "step": 342
+    },
+    {
+      "epoch": 5.94,
+      "learning_rate": 1.0056912603149229e-05,
+      "loss": 0.0469,
+      "step": 343
+    },
+    {
+      "epoch": 5.96,
+      "learning_rate": 1e-05,
+      "loss": 0.0479,
+      "step": 344
+    },
+    {
+      "epoch": 5.97,
+      "learning_rate": 9.943087396850773e-06,
+      "loss": 0.047,
+      "step": 345
+    },
+    {
+      "epoch": 5.99,
+      "learning_rate": 9.886176637140959e-06,
+      "loss": 0.046,
+      "step": 346
+    },
+    {
+      "epoch": 6.01,
+      "learning_rate": 9.829269564250254e-06,
+      "loss": 0.0313,
+      "step": 347
+    },
+    {
+      "epoch": 6.03,
+      "learning_rate": 9.772368021438943e-06,
+      "loss": 0.0222,
+      "step": 348
+    },
+    {
+      "epoch": 6.04,
+      "learning_rate": 9.715473851788187e-06,
+      "loss": 0.0215,
+      "step": 349
+    },
+    {
+      "epoch": 6.06,
+      "learning_rate": 9.658588898140322e-06,
+      "loss": 0.0218,
+      "step": 350
+    },
+    {
+      "epoch": 6.08,
+      "learning_rate": 9.601715003039174e-06,
+      "loss": 0.0204,
+      "step": 351
+    },
+    {
+      "epoch": 6.1,
+      "learning_rate": 9.544854008670366e-06,
+      "loss": 0.019,
+      "step": 352
+    },
+    {
+      "epoch": 6.11,
+      "learning_rate": 9.488007756801672e-06,
+      "loss": 0.0204,
+      "step": 353
+    },
+    {
+      "epoch": 6.13,
+      "learning_rate": 9.431178088723334e-06,
+      "loss": 0.0194,
+      "step": 354
+    },
+    {
+      "epoch": 6.15,
+      "learning_rate": 9.374366845188441e-06,
+      "loss": 0.0186,
+      "step": 355
+    },
+    {
+      "epoch": 6.16,
+      "learning_rate": 9.317575866353293e-06,
+      "loss": 0.0175,
+      "step": 356
+    },
+    {
+      "epoch": 6.18,
+      "learning_rate": 9.260806991717802e-06,
+      "loss": 0.0192,
+      "step": 357
+    },
+    {
+      "epoch": 6.2,
+      "learning_rate": 9.204062060065915e-06,
+      "loss": 0.0171,
+      "step": 358
+    },
+    {
+      "epoch": 6.22,
+      "learning_rate": 9.14734290940604e-06,
+      "loss": 0.0165,
+      "step": 359
+    },
+    {
+      "epoch": 6.23,
+      "learning_rate": 9.090651376911532e-06,
+      "loss": 0.0196,
+      "step": 360
+    },
+    {
+      "epoch": 6.25,
+      "learning_rate": 9.033989298861162e-06,
+      "loss": 0.0185,
+      "step": 361
+    },
+    {
+      "epoch": 6.27,
+      "learning_rate": 8.977358510579658e-06,
+      "loss": 0.0181,
+      "step": 362
+    },
+    {
+      "epoch": 6.29,
+      "learning_rate": 8.920760846378248e-06,
+      "loss": 0.0191,
+      "step": 363
+    },
+    {
+      "epoch": 6.3,
+      "learning_rate": 8.86419813949525e-06,
+      "loss": 0.0193,
+      "step": 364
+    },
+    {
+      "epoch": 6.32,
+      "learning_rate": 8.807672222036692e-06,
+      "loss": 0.0186,
+      "step": 365
+    },
+    {
+      "epoch": 6.34,
+      "learning_rate": 8.751184924916954e-06,
+      "loss": 0.0184,
+      "step": 366
+    },
+    {
+      "epoch": 6.35,
+      "learning_rate": 8.694738077799487e-06,
+      "loss": 0.0183,
+      "step": 367
+    },
+    {
+      "epoch": 6.37,
+      "learning_rate": 8.638333509037537e-06,
+      "loss": 0.0183,
+      "step": 368
+    },
+    {
+      "epoch": 6.39,
+      "learning_rate": 8.58197304561492e-06,
+      "loss": 0.0201,
+      "step": 369
+    },
+    {
+      "epoch": 6.41,
+      "learning_rate": 8.525658513086857e-06,
+      "loss": 0.0183,
+      "step": 370
+    },
+    {
+      "epoch": 6.42,
+      "learning_rate": 8.469391735520824e-06,
+      "loss": 0.0185,
+      "step": 371
+    },
+    {
+      "epoch": 6.44,
+      "learning_rate": 8.413174535437486e-06,
+      "loss": 0.0178,
+      "step": 372
+    },
+    {
+      "epoch": 6.46,
+      "learning_rate": 8.357008733751664e-06,
+      "loss": 0.0184,
+      "step": 373
+    },
+    {
+      "epoch": 6.48,
+      "learning_rate": 8.300896149713334e-06,
+      "loss": 0.0176,
+      "step": 374
+    },
+    {
+      "epoch": 6.49,
+      "learning_rate": 8.244838600848727e-06,
+      "loss": 0.0169,
+      "step": 375
+    },
+    {
+      "epoch": 6.51,
+      "learning_rate": 8.188837902901441e-06,
+      "loss": 0.0182,
+      "step": 376
+    },
+    {
+      "epoch": 6.53,
+      "learning_rate": 8.132895869773638e-06,
+      "loss": 0.0199,
+      "step": 377
+    },
+    {
+      "epoch": 6.55,
+      "learning_rate": 8.077014313467274e-06,
+      "loss": 0.0179,
+      "step": 378
+    },
+    {
+      "epoch": 6.56,
+      "learning_rate": 8.021195044025432e-06,
+      "loss": 0.0154,
+      "step": 379
+    },
+    {
+      "epoch": 6.58,
+      "learning_rate": 7.965439869473664e-06,
+      "loss": 0.0173,
+      "step": 380
+    },
+    {
+      "epoch": 6.6,
+      "learning_rate": 7.909750595761459e-06,
+      "loss": 0.0178,
+      "step": 381
+    },
+    {
+      "epoch": 6.61,
+      "learning_rate": 7.854129026703716e-06,
+      "loss": 0.0175,
+      "step": 382
+    },
+    {
+      "epoch": 6.63,
+      "learning_rate": 7.798576963922347e-06,
+      "loss": 0.0204,
+      "step": 383
+    },
+    {
+      "epoch": 6.65,
+      "learning_rate": 7.743096206787894e-06,
+      "loss": 0.0164,
+      "step": 384
+    },
+    {
+      "epoch": 6.67,
+      "learning_rate": 7.687688552361272e-06,
+      "loss": 0.0168,
+      "step": 385
+    },
+    {
+      "epoch": 6.68,
+      "learning_rate": 7.632355795335533e-06,
+      "loss": 0.0171,
+      "step": 386
+    },
+    {
+      "epoch": 6.7,
+      "learning_rate": 7.577099727977762e-06,
+      "loss": 0.018,
+      "step": 387
+    },
+    {
+      "epoch": 6.72,
+      "learning_rate": 7.521922140071003e-06,
+      "loss": 0.019,
+      "step": 388
+    },
+    {
+      "epoch": 6.74,
+      "learning_rate": 7.466824818856296e-06,
+      "loss": 0.0174,
+      "step": 389
+    },
+    {
+      "epoch": 6.75,
+      "learning_rate": 7.411809548974792e-06,
+      "loss": 0.0189,
+      "step": 390
+    },
+    {
+      "epoch": 6.77,
+      "learning_rate": 7.356878112409936e-06,
+      "loss": 0.0194,
+      "step": 391
+    },
+    {
+      "epoch": 6.79,
+      "learning_rate": 7.3020322884297565e-06,
+      "loss": 0.0163,
+      "step": 392
+    },
+    {
+      "epoch": 6.81,
+      "learning_rate": 7.2472738535292295e-06,
+      "loss": 0.0175,
+      "step": 393
+    },
+    {
+      "epoch": 6.82,
+      "learning_rate": 7.192604581372727e-06,
+      "loss": 0.0174,
+      "step": 394
+    },
+    {
+      "epoch": 6.84,
+      "learning_rate": 7.1380262427365885e-06,
+      "loss": 0.0173,
+      "step": 395
+    },
+    {
+      "epoch": 6.86,
+      "learning_rate": 7.0835406054517505e-06,
+      "loss": 0.0181,
+      "step": 396
+    },
+    {
+      "epoch": 6.87,
+      "learning_rate": 7.0291494343464896e-06,
+      "loss": 0.0163,
+      "step": 397
+    },
+    {
+      "epoch": 6.89,
+      "learning_rate": 6.974854491189243e-06,
+      "loss": 0.0191,
+      "step": 398
+    },
+    {
+      "epoch": 6.91,
+      "learning_rate": 6.920657534631573e-06,
+      "loss": 0.0175,
+      "step": 399
+    },
+    {
+      "epoch": 6.93,
+      "learning_rate": 6.866560320151179e-06,
+      "loss": 0.0174,
+      "step": 400
+    },
+    {
+      "epoch": 6.94,
+      "learning_rate": 6.812564599995042e-06,
+      "loss": 0.0167,
+      "step": 401
+    },
+    {
+      "epoch": 6.96,
+      "learning_rate": 6.758672123122675e-06,
+      "loss": 0.0185,
+      "step": 402
+    },
+    {
+      "epoch": 6.98,
+      "learning_rate": 6.704884635149467e-06,
+      "loss": 0.0167,
+      "step": 403
+    },
+    {
+      "epoch": 7.0,
+      "learning_rate": 6.651203878290139e-06,
+      "loss": 0.0174,
+      "step": 404
+    },
+    {
+      "epoch": 7.01,
+      "learning_rate": 6.597631591302319e-06,
+      "loss": 0.0104,
+      "step": 405
+    },
+    {
+      "epoch": 7.03,
+      "learning_rate": 6.544169509430219e-06,
+      "loss": 0.0085,
+      "step": 406
+    },
+    {
+      "epoch": 7.05,
+      "learning_rate": 6.490819364348434e-06,
+      "loss": 0.0089,
+      "step": 407
+    },
+    {
+      "epoch": 7.06,
+      "learning_rate": 6.437582884105835e-06,
+      "loss": 0.0092,
+      "step": 408
+    },
+    {
+      "epoch": 7.08,
+      "learning_rate": 6.384461793069616e-06,
+      "loss": 0.0083,
+      "step": 409
+    },
+    {
+      "epoch": 7.1,
+      "learning_rate": 6.331457811869437e-06,
+      "loss": 0.0088,
+      "step": 410
+    },
+    {
+      "epoch": 7.12,
+      "learning_rate": 6.278572657341682e-06,
+      "loss": 0.0084,
+      "step": 411
+    },
+    {
+      "epoch": 7.13,
+      "learning_rate": 6.225808042473857e-06,
+      "loss": 0.0089,
+      "step": 412
+    },
+    {
+      "epoch": 7.15,
+      "learning_rate": 6.173165676349103e-06,
+      "loss": 0.0078,
+      "step": 413
+    },
+    {
+      "epoch": 7.17,
+      "learning_rate": 6.120647264090839e-06,
+      "loss": 0.0073,
+      "step": 414
+    },
+    {
+      "epoch": 7.19,
+      "learning_rate": 6.0682545068075315e-06,
+      "loss": 0.0088,
+      "step": 415
+    },
+    {
+      "epoch": 7.2,
+      "learning_rate": 6.015989101537586e-06,
+      "loss": 0.0071,
+      "step": 416
+    },
+    {
+      "epoch": 7.22,
+      "learning_rate": 5.963852741194397e-06,
+      "loss": 0.0085,
+      "step": 417
+    },
+    {
+      "epoch": 7.24,
+      "learning_rate": 5.911847114511497e-06,
+      "loss": 0.0079,
+      "step": 418
+    },
+    {
+      "epoch": 7.26,
+      "learning_rate": 5.859973905987866e-06,
+      "loss": 0.0077,
+      "step": 419
+    },
+    {
+      "epoch": 7.27,
+      "learning_rate": 5.8082347958333625e-06,
+      "loss": 0.0085,
+      "step": 420
+    },
+    {
+      "epoch": 7.29,
+      "learning_rate": 5.756631459914302e-06,
+      "loss": 0.008,
+      "step": 421
+    },
+    {
+      "epoch": 7.31,
+      "learning_rate": 5.7051655696991825e-06,
+      "loss": 0.0072,
+      "step": 422
+    },
+    {
+      "epoch": 7.32,
+      "learning_rate": 5.653838792204538e-06,
+      "loss": 0.0072,
+      "step": 423
+    },
+    {
+      "epoch": 7.34,
+      "learning_rate": 5.602652789940941e-06,
+      "loss": 0.0075,
+      "step": 424
+    },
+    {
+      "epoch": 7.36,
+      "learning_rate": 5.55160922085916e-06,
+      "loss": 0.0086,
+      "step": 425
+    },
+    {
+      "epoch": 7.38,
+      "learning_rate": 5.50070973829644e-06,
+      "loss": 0.0064,
+      "step": 426
+    },
+    {
+      "epoch": 7.39,
+      "learning_rate": 5.449955990922973e-06,
+      "loss": 0.0087,
+      "step": 427
+    },
+    {
+      "epoch": 7.41,
+      "learning_rate": 5.399349622688479e-06,
+      "loss": 0.0077,
+      "step": 428
+    },
+    {
+      "epoch": 7.43,
+      "learning_rate": 5.348892272768972e-06,
+      "loss": 0.0071,
+      "step": 429
+    },
+    {
+      "epoch": 7.45,
+      "learning_rate": 5.298585575513649e-06,
+      "loss": 0.0073,
+      "step": 430
+    },
+    {
+      "epoch": 7.46,
+      "learning_rate": 5.248431160391963e-06,
+      "loss": 0.0067,
+      "step": 431
+    },
+    {
+      "epoch": 7.48,
+      "learning_rate": 5.198430651940846e-06,
+      "loss": 0.0072,
+      "step": 432
+    },
+    {
+      "epoch": 7.5,
+      "learning_rate": 5.148585669712074e-06,
+      "loss": 0.0071,
+      "step": 433
+    },
+    {
+      "epoch": 7.52,
+      "learning_rate": 5.098897828219831e-06,
+      "loss": 0.0078,
+      "step": 434
+    },
+    {
+      "epoch": 7.53,
+      "learning_rate": 5.049368736888391e-06,
+      "loss": 0.0075,
+      "step": 435
+    },
+    {
+      "epoch": 7.55,
+      "learning_rate": 5.000000000000003e-06,
+      "loss": 0.0068,
+      "step": 436
+    },
+    {
+      "epoch": 7.57,
+      "learning_rate": 4.950793216642923e-06,
+      "loss": 0.0075,
+      "step": 437
+    },
+    {
+      "epoch": 7.58,
+      "learning_rate": 4.901749980659617e-06,
+      "loss": 0.0072,
+      "step": 438
+    },
+    {
+      "epoch": 7.6,
+      "learning_rate": 4.852871880595137e-06,
+      "loss": 0.0063,
+      "step": 439
+    },
+    {
+      "epoch": 7.62,
+      "learning_rate": 4.804160499645667e-06,
+      "loss": 0.0073,
+      "step": 440
+    },
+    {
+      "epoch": 7.64,
+      "learning_rate": 4.755617415607232e-06,
+      "loss": 0.0075,
+      "step": 441
+    },
+    {
+      "epoch": 7.65,
+      "learning_rate": 4.7072442008246135e-06,
+      "loss": 0.0075,
+      "step": 442
+    },
+    {
+      "epoch": 7.67,
+      "learning_rate": 4.659042422140399e-06,
+      "loss": 0.0079,
+      "step": 443
+    },
+    {
+      "epoch": 7.69,
+      "learning_rate": 4.611013640844245e-06,
+      "loss": 0.0067,
+      "step": 444
+    },
+    {
+      "epoch": 7.71,
+      "learning_rate": 4.5631594126223e-06,
+      "loss": 0.0072,
+      "step": 445
+    },
+    {
+      "epoch": 7.72,
+      "learning_rate": 4.515481287506811e-06,
+      "loss": 0.0081,
+      "step": 446
+    },
+    {
+      "epoch": 7.74,
+      "learning_rate": 4.4679808098259295e-06,
+      "loss": 0.0071,
+      "step": 447
+    },
+    {
+      "epoch": 7.76,
+      "learning_rate": 4.420659518153667e-06,
+      "loss": 0.0075,
+      "step": 448
+    },
+    {
+      "epoch": 7.77,
+      "learning_rate": 4.37351894526009e-06,
+      "loss": 0.0073,
+      "step": 449
+    },
+    {
+      "epoch": 7.79,
+      "learning_rate": 4.326560618061639e-06,
+      "loss": 0.007,
+      "step": 450
+    },
+    {
+      "epoch": 7.81,
+      "learning_rate": 4.279786057571703e-06,
+      "loss": 0.0079,
+      "step": 451
+    },
+    {
+      "epoch": 7.83,
+      "learning_rate": 4.2331967788513295e-06,
+      "loss": 0.0075,
+      "step": 452
+    },
+    {
+      "epoch": 7.84,
+      "learning_rate": 4.186794290960162e-06,
+      "loss": 0.0067,
+      "step": 453
+    },
+    {
+      "epoch": 7.86,
+      "learning_rate": 4.140580096907554e-06,
+      "loss": 0.0058,
+      "step": 454
+    },
+    {
+      "epoch": 7.88,
+      "learning_rate": 4.094555693603891e-06,
+      "loss": 0.0076,
+      "step": 455
+    },
+    {
+      "epoch": 7.9,
+      "learning_rate": 4.048722571812105e-06,
+      "loss": 0.0075,
+      "step": 456
+    },
+    {
+      "epoch": 7.91,
+      "learning_rate": 4.003082216099374e-06,
+      "loss": 0.007,
+      "step": 457
+    },
+    {
+      "epoch": 7.93,
+      "learning_rate": 3.957636104789056e-06,
+      "loss": 0.0065,
+      "step": 458
+    },
+    {
+      "epoch": 7.95,
+      "learning_rate": 3.912385709912794e-06,
+      "loss": 0.0077,
+      "step": 459
+    },
+    {
+      "epoch": 7.97,
+      "learning_rate": 3.867332497162836e-06,
+      "loss": 0.0072,
+      "step": 460
+    },
+    {
+      "epoch": 7.98,
+      "learning_rate": 3.822477925844564e-06,
+      "loss": 0.0077,
+      "step": 461
+    },
+    {
+      "epoch": 8.0,
+      "learning_rate": 3.777823448829224e-06,
+      "loss": 0.0066,
+      "step": 462
+    },
+    {
+      "epoch": 8.02,
+      "learning_rate": 3.7333705125068576e-06,
+      "loss": 0.0041,
+      "step": 463
+    },
+    {
+      "epoch": 8.03,
+      "learning_rate": 3.689120556739475e-06,
+      "loss": 0.0048,
+      "step": 464
+    },
+    {
+      "epoch": 8.05,
+      "learning_rate": 3.6450750148143886e-06,
+      "loss": 0.0035,
+      "step": 465
+    },
+    {
+      "epoch": 8.07,
+      "learning_rate": 3.601235313397813e-06,
+      "loss": 0.0045,
+      "step": 466
+    },
+    {
+      "epoch": 8.09,
+      "learning_rate": 3.557602872488638e-06,
+      "loss": 0.0041,
+      "step": 467
+    },
+    {
+      "epoch": 8.1,
+      "learning_rate": 3.5141791053724405e-06,
+      "loss": 0.0042,
+      "step": 468
+    },
+    {
+      "epoch": 8.12,
+      "learning_rate": 3.470965418575708e-06,
+      "loss": 0.004,
+      "step": 469
+    },
+    {
+      "epoch": 8.14,
+      "learning_rate": 3.4279632118202744e-06,
+      "loss": 0.0039,
+      "step": 470
+    },
+    {
+      "epoch": 8.16,
+      "learning_rate": 3.385173877977991e-06,
+      "loss": 0.0039,
+      "step": 471
+    },
+    {
+      "epoch": 8.17,
+      "learning_rate": 3.342598803025595e-06,
+      "loss": 0.004,
+      "step": 472
+    },
+    {
+      "epoch": 8.19,
+      "learning_rate": 3.3002393659998357e-06,
+      "loss": 0.0039,
+      "step": 473
+    },
+    {
+      "epoch": 8.21,
+      "learning_rate": 3.258096938952796e-06,
+      "loss": 0.0036,
+      "step": 474
+    },
+    {
+      "epoch": 8.23,
+      "learning_rate": 3.2161728869074517e-06,
+      "loss": 0.0033,
+      "step": 475
+    },
+    {
+      "epoch": 8.24,
+      "learning_rate": 3.174468567813461e-06,
+      "loss": 0.0037,
+      "step": 476
+    },
+    {
+      "epoch": 8.26,
+      "learning_rate": 3.132985332503167e-06,
+      "loss": 0.0031,
+      "step": 477
+    },
+    {
+      "epoch": 8.28,
+      "learning_rate": 3.091724524647861e-06,
+      "loss": 0.0043,
+      "step": 478
+    },
+    {
+      "epoch": 8.29,
+      "learning_rate": 3.050687480714256e-06,
+      "loss": 0.0034,
+      "step": 479
+    },
+    {
+      "epoch": 8.31,
+      "learning_rate": 3.009875529921181e-06,
+      "loss": 0.0037,
+      "step": 480
+    },
+    {
+      "epoch": 8.33,
+      "learning_rate": 2.9692899941965493e-06,
+      "loss": 0.004,
+      "step": 481
+    },
+    {
+      "epoch": 8.35,
+      "learning_rate": 2.9289321881345257e-06,
+      "loss": 0.0039,
+      "step": 482
+    },
+    {
+      "epoch": 8.36,
+      "learning_rate": 2.8888034189529524e-06,
+      "loss": 0.0037,
+      "step": 483
+    },
+    {
+      "epoch": 8.38,
+      "learning_rate": 2.8489049864510053e-06,
+      "loss": 0.004,
+      "step": 484
+    },
+    {
+      "epoch": 8.4,
+      "learning_rate": 2.809238182967092e-06,
+      "loss": 0.0042,
+      "step": 485
+    },
+    {
+      "epoch": 8.42,
+      "learning_rate": 2.769804293336994e-06,
+      "loss": 0.0039,
+      "step": 486
+    },
+    {
+      "epoch": 8.43,
+      "learning_rate": 2.730604594852243e-06,
+      "loss": 0.004,
+      "step": 487
+    },
+    {
+      "epoch": 8.45,
+      "learning_rate": 2.691640357218759e-06,
+      "loss": 0.0037,
+      "step": 488
+    },
+    {
+      "epoch": 8.47,
+      "learning_rate": 2.6529128425157226e-06,
+      "loss": 0.0043,
+      "step": 489
+    },
+    {
+      "epoch": 8.48,
+      "learning_rate": 2.6144233051546797e-06,
+      "loss": 0.0041,
+      "step": 490
+    },
+    {
+      "epoch": 8.5,
+      "learning_rate": 2.576172991838933e-06,
+      "loss": 0.0036,
+      "step": 491
+    },
+    {
+      "epoch": 8.52,
+      "learning_rate": 2.5381631415231455e-06,
+      "loss": 0.0036,
+      "step": 492
+    },
+    {
+      "epoch": 8.54,
+      "learning_rate": 2.5003949853732135e-06,
+      "loss": 0.0036,
+      "step": 493
+    },
+    {
+      "epoch": 8.55,
+      "learning_rate": 2.4628697467263916e-06,
+      "loss": 0.0042,
+      "step": 494
+    },
+    {
+      "epoch": 8.57,
+      "learning_rate": 2.425588641051656e-06,
+      "loss": 0.0038,
+      "step": 495
+    },
+    {
+      "epoch": 8.59,
+      "learning_rate": 2.388552875910354e-06,
+      "loss": 0.004,
+      "step": 496
+    },
+    {
+      "epoch": 8.61,
+      "learning_rate": 2.351763650917074e-06,
+      "loss": 0.0038,
+      "step": 497
+    },
+    {
+      "epoch": 8.62,
+      "learning_rate": 2.315222157700797e-06,
+      "loss": 0.0037,
+      "step": 498
+    },
+    {
+      "epoch": 8.64,
+      "learning_rate": 2.2789295798662967e-06,
+      "loss": 0.004,
+      "step": 499
+    },
+    {
+      "epoch": 8.66,
+      "learning_rate": 2.2428870929558012e-06,
+      "loss": 0.0039,
+      "step": 500
+    },
+    {
+      "epoch": 8.68,
+      "learning_rate": 2.207095864410919e-06,
+      "loss": 0.004,
+      "step": 501
+    },
+    {
+      "epoch": 8.69,
+      "learning_rate": 2.171557053534814e-06,
+      "loss": 0.0043,
+      "step": 502
+    },
+    {
+      "epoch": 8.71,
+      "learning_rate": 2.1362718114546777e-06,
+      "loss": 0.0038,
+      "step": 503
+    },
+    {
+      "epoch": 8.73,
+      "learning_rate": 2.101241281084416e-06,
+      "loss": 0.0033,
+      "step": 504
+    },
+    {
+      "epoch": 8.74,
+      "learning_rate": 2.0664665970876496e-06,
+      "loss": 0.004,
+      "step": 505
+    },
+    {
+      "epoch": 8.76,
+      "learning_rate": 2.0319488858409552e-06,
+      "loss": 0.0037,
+      "step": 506
+    },
+    {
+      "epoch": 8.78,
+      "learning_rate": 1.997689265397377e-06,
+      "loss": 0.0038,
+      "step": 507
+    },
+    {
+      "epoch": 8.8,
+      "learning_rate": 1.963688845450218e-06,
+      "loss": 0.0041,
+      "step": 508
+    },
+    {
+      "epoch": 8.81,
+      "learning_rate": 1.929948727297096e-06,
+      "loss": 0.0042,
+      "step": 509
+    },
+    {
+      "epoch": 8.83,
+      "learning_rate": 1.8964700038042628e-06,
+      "loss": 0.0032,
+      "step": 510
+    },
+    {
+      "epoch": 8.85,
+      "learning_rate": 1.8632537593712186e-06,
+      "loss": 0.0038,
+      "step": 511
+    },
+    {
+      "epoch": 8.87,
+      "learning_rate": 1.8303010698955803e-06,
+      "loss": 0.003,
+      "step": 512
+    },
+    {
+      "epoch": 8.88,
+      "learning_rate": 1.7976130027382332e-06,
+      "loss": 0.0035,
+      "step": 513
+    },
+    {
+      "epoch": 8.9,
+      "learning_rate": 1.76519061668876e-06,
+      "loss": 0.0037,
+      "step": 514
+    },
+    {
+      "epoch": 8.92,
+      "learning_rate": 1.7330349619311415e-06,
+      "loss": 0.0038,
+      "step": 515
+    },
+    {
+      "epoch": 8.94,
+      "learning_rate": 1.7011470800097496e-06,
+      "loss": 0.0042,
+      "step": 516
+    },
+    {
+      "epoch": 8.95,
+      "learning_rate": 1.6695280037955953e-06,
+      "loss": 0.0031,
+      "step": 517
+    },
+    {
+      "epoch": 8.97,
+      "learning_rate": 1.638178757452894e-06,
+      "loss": 0.0035,
+      "step": 518
+    },
+    {
+      "epoch": 8.99,
+      "learning_rate": 1.6071003564058697e-06,
+      "loss": 0.0041,
+      "step": 519
+    },
+    {
+      "epoch": 9.0,
+      "learning_rate": 1.5762938073058853e-06,
+      "loss": 0.003,
+      "step": 520
+    },
+    {
+      "epoch": 9.02,
+      "learning_rate": 1.5457601079988226e-06,
+      "loss": 0.0025,
+      "step": 521
+    },
+    {
+      "epoch": 9.04,
+      "learning_rate": 1.5155002474927683e-06,
+      "loss": 0.0025,
+      "step": 522
+    },
+    {
+      "epoch": 9.06,
+      "learning_rate": 1.4855152059259737e-06,
+      "loss": 0.0026,
+      "step": 523
+    },
+    {
+      "epoch": 9.07,
+      "learning_rate": 1.4558059545351144e-06,
+      "loss": 0.0033,
+      "step": 524
+    },
+    {
+      "epoch": 9.09,
+      "learning_rate": 1.4263734556238262e-06,
+      "loss": 0.0027,
+      "step": 525
+    },
+    {
+      "epoch": 9.11,
+      "learning_rate": 1.397218662531532e-06,
+      "loss": 0.0028,
+      "step": 526
+    },
+    {
+      "epoch": 9.13,
+      "learning_rate": 1.3683425196025734e-06,
+      "loss": 0.0029,
+      "step": 527
+    },
+    {
+      "epoch": 9.14,
+      "learning_rate": 1.339745962155613e-06,
+      "loss": 0.0025,
+      "step": 528
+    },
+    {
+      "epoch": 9.16,
+      "learning_rate": 1.3114299164533451e-06,
+      "loss": 0.0023,
+      "step": 529
+    },
+    {
+      "epoch": 9.18,
+      "learning_rate": 1.2833952996724864e-06,
+      "loss": 0.0034,
+      "step": 530
+    },
+    {
+      "epoch": 9.19,
+      "learning_rate": 1.2556430198740776e-06,
+      "loss": 0.003,
+      "step": 531
+    },
+    {
+      "epoch": 9.21,
+      "learning_rate": 1.2281739759740575e-06,
+      "loss": 0.0024,
+      "step": 532
+    },
+    {
+      "epoch": 9.23,
+      "learning_rate": 1.2009890577141625e-06,
+      "loss": 0.0022,
+      "step": 533
+    },
+    {
+      "epoch": 9.25,
+      "learning_rate": 1.1740891456330894e-06,
+      "loss": 0.0026,
+      "step": 534
+    },
+    {
+      "epoch": 9.26,
+      "learning_rate": 1.1474751110379933e-06,
+      "loss": 0.0025,
+      "step": 535
+    },
+    {
+      "epoch": 9.28,
+      "learning_rate": 1.121147815976248e-06,
+      "loss": 0.0028,
+      "step": 536
+    },
+    {
+      "epoch": 9.3,
+      "learning_rate": 1.095108113207537e-06,
+      "loss": 0.0028,
+      "step": 537
+    },
+    {
+      "epoch": 9.32,
+      "learning_rate": 1.0693568461762238e-06,
+      "loss": 0.003,
+      "step": 538
+    },
+    {
+      "epoch": 9.33,
+      "learning_rate": 1.0438948489840327e-06,
+      "loss": 0.003,
+      "step": 539
+    },
+    {
+      "epoch": 9.35,
+      "learning_rate": 1.01872294636304e-06,
+      "loss": 0.0029,
+      "step": 540
+    },
+    {
+      "epoch": 9.37,
+      "learning_rate": 9.93841953648944e-07,
+      "loss": 0.0025,
+      "step": 541
+    },
+    {
+      "epoch": 9.39,
+      "learning_rate": 9.692526767546727e-07,
+      "loss": 0.0028,
+      "step": 542
+    },
+    {
+      "epoch": 9.4,
+      "learning_rate": 9.449559121442731e-07,
+      "loss": 0.0028,
+      "step": 543
+    },
+    {
+      "epoch": 9.42,
+      "learning_rate": 9.209524468071096e-07,
+      "loss": 0.0029,
+      "step": 544
+    },
+    {
+      "epoch": 9.44,
+      "learning_rate": 8.972430582323788e-07,
+      "loss": 0.0025,
+      "step": 545
+    },
+    {
+      "epoch": 9.45,
+      "learning_rate": 8.738285143839198e-07,
+      "loss": 0.0023,
+      "step": 546
+    },
+    {
+      "epoch": 9.47,
+      "learning_rate": 8.507095736753435e-07,
+      "loss": 0.0031,
+      "step": 547
+    },
+    {
+      "epoch": 9.49,
+      "learning_rate": 8.278869849454718e-07,
+      "loss": 0.0025,
+      "step": 548
+    },
+    {
+      "epoch": 9.51,
+      "learning_rate": 8.053614874340665e-07,
+      "loss": 0.0029,
+      "step": 549
+    },
+    {
+      "epoch": 9.52,
+      "learning_rate": 7.831338107579056e-07,
+      "loss": 0.0025,
+      "step": 550
+    }
+  ],
+  "max_steps": 570,
+  "num_train_epochs": 10,
+  "total_flos": 4.2387842097888625e+18,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0da89e30e3909c1102a0ac1bed23fb4da68f88e1d82ed81b225d0831775f5048
+size 5435

zero_to_fp32.py

@@ -0,0 +1,578 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage == 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dicts.append(torch.load(f, map_location=device))
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage == 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage == 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage == 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file)